Changing a Decision - Behavioral Mathematics for Game AI - page 413

Exactly how we apply this factor would depend largely on how we constructed
the decision algorithm in the first place. However, it is usually a good idea to bias a
distant objective pessimistically. For example, we would not serve ourselves well if
we assumed that, by the time we arrived, the enemy will have graciously removed
all their defenders.
In that respect, another method of biasing our decision is to add a pessimistic,
time-based value to the strength of the opponents in the city. Imagine flipping the
curve in Figure 15.2 upside down. Instead of a confidence factor moving downward
from 1.0 toward 0, we could create a multiplier that moves up from 1.0. By using
this value as a coefficient for the defenses, we are artificially inflating our estimate
of the defense capabilities of the enemy. This could represent the possibility that
they will see us coming and have time to move units into the city, for example. The
result is that we will be less likely to consider that option as viable.
On the other hand, we must also take care to not paint such a bleak picture of
all distant targets that our fleet despondently remains at anchor and doesn't do any-
thing . Remember, much of the time we are using the information to bias an other-
wise equal decision one way or another… not negate the whole process of action
entirely.
The Value of Thoroughness
Not all of the considerations on whether or not to switch from one task to another
are based on time, of course. Another significant factor reoccurs in many different
types of decision processes—completing the task at hand. Unfortunately, one of the
side effects of the task monitoring that we discussed at the beginning of the chapter
is that our agent may be tempted to abandon what it was doing to perform what it
believes is a higher priority task. While that is certainly laudable, and is a necessary
part of a proper decision framework, we don't want our agents simply abandoning
what they were doing because something else comes up. We would often end up
with many partially finished projects and nothing that ever gets completed .
One simple illustration of this effect is a slight modification of our fleet example.
Imagine that our little armada actually began to attack Dudetown rather than
changing its goal before the assault. However, just as the last battered and bleeding
defenders were about to fall, we received word that the distant Suckerville was un-
defended. If we leave at that moment and don't finish the sacking of Dudetown, we
give the residents time to rebuild, the military time to regroup and heal, and per-
haps even allow time for reinforcements to arrive. Even if we are only gone for a few
turns, we may be right back where we started in our quest to eradicate Dudetown
of all Dudes. What did we accomplish by our initial attack?